{
 "version": 1,
 "arms": [
  {
   "arm": "grpo_gs8_seed0",
   "base_checkpoint": "Qwen/Qwen3.5-2B",
   "init_checkpoint": "tinker://66bc785a6dd2ae0f72c6707928dfb24b/weights/run_cbb327f6ab-init",
   "init_method": "create_lora_training_client(rank=32; server rejects per-model seeds, so seed 0 not applied) -> save_state -> create_training_client_from_state (fresh optimizer state)",
   "adapter_sha256": null,
   "evidence_ref": "out/restoration.json",
   "before_first_update": true,
   "restore_count": 1,
   "optimizer_checks": [
    {
     "event": "restore_from_init_state",
     "expected": {
      "family": "adam",
      "lr": 4e-05,
      "beta1": 0.9,
      "beta2": 0.95,
      "eps": 1e-08,
      "state": "fresh"
     },
     "actual": {
      "type": "adamw"
     },
     "source": "TrainingClient.get_info().optimizer_config (server)",
     "verified": null
    }
   ],
   "initial_outputs": {
    "fixed_batch": "first 8 GSM8K train rows (question+answer), cross_entropy forward",
    "fixed_batch_sha256": "48e2999547b6e6029b0feb6658a241a32ce80f73ec1ce8407285769808f0abee",
    "logprobs_sha256": "5ac755852ee024e0bbe1b3d178b57520bf33a99d2b08981579fb8c409d173e87",
    "max_abs_diff_vs_init_client": 0.0,
    "init_client_logprobs_sha256": "5ac755852ee024e0bbe1b3d178b57520bf33a99d2b08981579fb8c409d173e87"
   },
   "server_info": {
    "model_name": "Qwen/Qwen3.5-2B",
    "lora_rank": 32,
    "is_lora": true,
    "optimizer_config": {
     "type": "adamw"
    }
   }
  },
  {
   "arm": "gs1_seed0",
   "base_checkpoint": "Qwen/Qwen3.5-2B",
   "init_checkpoint": "tinker://66bc785a6dd2ae0f72c6707928dfb24b/weights/run_cbb327f6ab-init",
   "init_method": "create_lora_training_client(rank=32; server rejects per-model seeds, so seed 0 not applied) -> save_state -> create_training_client_from_state (fresh optimizer state)",
   "adapter_sha256": null,
   "evidence_ref": "out/restoration.json",
   "before_first_update": true,
   "restore_count": 1,
   "optimizer_checks": [
    {
     "event": "restore_from_init_state",
     "expected": {
      "family": "adam",
      "lr": 4e-05,
      "beta1": 0.9,
      "beta2": 0.95,
      "eps": 1e-08,
      "state": "fresh"
     },
     "actual": {
      "type": "adamw"
     },
     "source": "TrainingClient.get_info().optimizer_config (server)",
     "verified": null
    }
   ],
   "initial_outputs": {
    "fixed_batch": "first 8 GSM8K train rows (question+answer), cross_entropy forward",
    "fixed_batch_sha256": "48e2999547b6e6029b0feb6658a241a32ce80f73ec1ce8407285769808f0abee",
    "logprobs_sha256": "5ac755852ee024e0bbe1b3d178b57520bf33a99d2b08981579fb8c409d173e87",
    "max_abs_diff_vs_init_client": 0.0,
    "init_client_logprobs_sha256": "5ac755852ee024e0bbe1b3d178b57520bf33a99d2b08981579fb8c409d173e87"
   },
   "server_info": {
    "model_name": "Qwen/Qwen3.5-2B",
    "lora_rank": 32,
    "is_lora": true,
    "optimizer_config": {
     "type": "adamw"
    }
   }
  }
 ]
}